Thought experiment says any model costing $1M+ to train should be open-sourced
ylecun · x · 2026-07-27
A repost of a thought experiment asking what would happen if every sufficiently expensive ML model had to be open-sourced immediately.
The proposal would require releasing:
- the model weights
- the code needed to run the model
- detailed training procedure information
- possibly the training data as well
The post frames “sufficiently expensive” as a model that cost about $1 million or more to train, and uses the idea to probe the implications for open models, competition, and regulation.
More from AGI Musings
- Archive catalogs decades of failed AI predictions and pushes back on inevitability claims — Dan_Jeffries1 · 2026-07-27
- Brands Pivot to Micro-Influencers Amid AI Job Displacement — davidfromkansas · 2026-07-27
- A regulatory-markets model could bridge fast-moving AI and slow legislation — ghadfield · 2026-07-27
- AI agents may need revocable licenses as insurers pull back from generative AI — ghadfield · 2026-07-27
- Tim Maudlin traces Einstein, EPR, and Bell’s theorem in a long physics lecture — TOEwithCurt · 2026-07-27
- A consciousness debate asks whether biological properties are really necessary — aran_nayebi · 2026-07-27